migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu-common.h"
  17 #include "migration.h"
  18 #include "monitor.h"
  19 #include "buffered_file.h"
  20 #include "sysemu.h"
  21 #include "block.h"
  22 #include "qemu_socket.h"
  23 #include "block-migration.h"
  24 #include "qmp-commands.h"
  25
  26 //#define DEBUG_MIGRATION
  27
  28 #ifdef DEBUG_MIGRATION
  29 #define DPRINTF(fmt, ...) \
  30     do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
  31 #else
  32 #define DPRINTF(fmt, ...) \
  33     do { } while (0)
  34 #endif
  35
  36 enum {
  37     MIG_STATE_ERROR,
  38     MIG_STATE_SETUP,
  39     MIG_STATE_CANCELLED,
  40     MIG_STATE_ACTIVE,
  41     MIG_STATE_COMPLETED,
  42 };
  43
  44 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  45
  46 /* Migration XBZRLE default cache size */
  47 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  48
  49 static NotifierList migration_state_notifiers =
  50     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  51
  52 /* When we add fault tolerance, we could have several
  53    migrations at once.  For now we don't need to add
  54    dynamic creation of migration */
  55
  56 MigrationState *migrate_get_current(void)
  57 {
  58     static MigrationState current_migration = {
  59         .state = MIG_STATE_SETUP,
  60         .bandwidth_limit = MAX_THROTTLE,
  61         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  62     };
  63
  64     return &current_migration;
  65 }
  66
  67 void qemu_start_incoming_migration(const char *uri, Error **errp)
  68 {
  69     const char *p;
  70
  71     if (strstart(uri, "tcp:", &p))
  72         tcp_start_incoming_migration(p, errp);
  73 #if !defined(WIN32)
  74     else if (strstart(uri, "exec:", &p))
  75         exec_start_incoming_migration(p, errp);
  76     else if (strstart(uri, "unix:", &p))
  77         unix_start_incoming_migration(p, errp);
  78     else if (strstart(uri, "fd:", &p))
  79         fd_start_incoming_migration(p, errp);
  80 #endif
  81     else {
  82         error_setg(errp, "unknown migration protocol: %s\n", uri);
  83     }
  84 }
  85
  86 static void process_incoming_migration_co(void *opaque)
  87 {
  88     QEMUFile *f = opaque;
  89     int ret;
  90
  91     ret = qemu_loadvm_state(f);
  92     qemu_set_fd_handler(qemu_get_fd(f), NULL, NULL, NULL);
  93     qemu_fclose(f);
  94     if (ret < 0) {
  95         fprintf(stderr, "load of migration failed\n");
  96         exit(0);
  97     }
  98     qemu_announce_self();
  99     DPRINTF("successfully loaded vm state\n");
 100
 101     bdrv_clear_incoming_migration_all();
 102     /* Make sure all file formats flush their mutable metadata */
 103     bdrv_invalidate_cache_all();
 104
 105     if (autostart) {
 106         vm_start();
 107     } else {
 108         runstate_set(RUN_STATE_PAUSED);
 109     }
 110 }
 111
 112 static void enter_migration_coroutine(void *opaque)
 113 {
 114     Coroutine *co = opaque;
 115     qemu_coroutine_enter(co, NULL);
 116 }
 117
 118 void process_incoming_migration(QEMUFile *f)
 119 {
 120     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 121     int fd = qemu_get_fd(f);
 122
 123     assert(fd != -1);
 124     socket_set_nonblock(fd);
 125     qemu_set_fd_handler(fd, enter_migration_coroutine, NULL, co);
 126     qemu_coroutine_enter(co, f);
 127 }
 128
 129 /* amount of nanoseconds we are willing to wait for migration to be down.
 130  * the choice of nanoseconds is because it is the maximum resolution that
 131  * get_clock() can achieve. It is an internal measure. All user-visible
 132  * units must be in seconds */
 133 static uint64_t max_downtime = 30000000;
 134
 135 uint64_t migrate_max_downtime(void)
 136 {
 137     return max_downtime;
 138 }
 139
 140 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 141 {
 142     MigrationCapabilityStatusList *head = NULL;
 143     MigrationCapabilityStatusList *caps;
 144     MigrationState *s = migrate_get_current();
 145     int i;
 146
 147     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 148         if (head == NULL) {
 149             head = g_malloc0(sizeof(*caps));
 150             caps = head;
 151         } else {
 152             caps->next = g_malloc0(sizeof(*caps));
 153             caps = caps->next;
 154         }
 155         caps->value =
 156             g_malloc(sizeof(*caps->value));
 157         caps->value->capability = i;
 158         caps->value->state = s->enabled_capabilities[i];
 159     }
 160
 161     return head;
 162 }
 163
 164 static void get_xbzrle_cache_stats(MigrationInfo *info)
 165 {
 166     if (migrate_use_xbzrle()) {
 167         info->has_xbzrle_cache = true;
 168         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 169         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 170         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 171         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 172         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 173         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 174     }
 175 }
 176
 177 MigrationInfo *qmp_query_migrate(Error **errp)
 178 {
 179     MigrationInfo *info = g_malloc0(sizeof(*info));
 180     MigrationState *s = migrate_get_current();
 181
 182     switch (s->state) {
 183     case MIG_STATE_SETUP:
 184         /* no migration has happened ever */
 185         break;
 186     case MIG_STATE_ACTIVE:
 187         info->has_status = true;
 188         info->status = g_strdup("active");
 189         info->has_total_time = true;
 190         info->total_time = qemu_get_clock_ms(rt_clock)
 191             - s->total_time;
 192         info->has_expected_downtime = true;
 193         info->expected_downtime = s->expected_downtime;
 194
 195         info->has_ram = true;
 196         info->ram = g_malloc0(sizeof(*info->ram));
 197         info->ram->transferred = ram_bytes_transferred();
 198         info->ram->remaining = ram_bytes_remaining();
 199         info->ram->total = ram_bytes_total();
 200         info->ram->duplicate = dup_mig_pages_transferred();
 201         info->ram->normal = norm_mig_pages_transferred();
 202         info->ram->normal_bytes = norm_mig_bytes_transferred();
 203         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 204
 205
 206         if (blk_mig_active()) {
 207             info->has_disk = true;
 208             info->disk = g_malloc0(sizeof(*info->disk));
 209             info->disk->transferred = blk_mig_bytes_transferred();
 210             info->disk->remaining = blk_mig_bytes_remaining();
 211             info->disk->total = blk_mig_bytes_total();
 212         }
 213
 214         get_xbzrle_cache_stats(info);
 215         break;
 216     case MIG_STATE_COMPLETED:
 217         get_xbzrle_cache_stats(info);
 218
 219         info->has_status = true;
 220         info->status = g_strdup("completed");
 221         info->total_time = s->total_time;
 222         info->has_downtime = true;
 223         info->downtime = s->downtime;
 224
 225         info->has_ram = true;
 226         info->ram = g_malloc0(sizeof(*info->ram));
 227         info->ram->transferred = ram_bytes_transferred();
 228         info->ram->remaining = 0;
 229         info->ram->total = ram_bytes_total();
 230         info->ram->duplicate = dup_mig_pages_transferred();
 231         info->ram->normal = norm_mig_pages_transferred();
 232         info->ram->normal_bytes = norm_mig_bytes_transferred();
 233         break;
 234     case MIG_STATE_ERROR:
 235         info->has_status = true;
 236         info->status = g_strdup("failed");
 237         break;
 238     case MIG_STATE_CANCELLED:
 239         info->has_status = true;
 240         info->status = g_strdup("cancelled");
 241         break;
 242     }
 243
 244     return info;
 245 }
 246
 247 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 248                                   Error **errp)
 249 {
 250     MigrationState *s = migrate_get_current();
 251     MigrationCapabilityStatusList *cap;
 252
 253     if (s->state == MIG_STATE_ACTIVE) {
 254         error_set(errp, QERR_MIGRATION_ACTIVE);
 255         return;
 256     }
 257
 258     for (cap = params; cap; cap = cap->next) {
 259         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 260     }
 261 }
 262
 263 /* shared migration helpers */
 264
 265 static int migrate_fd_cleanup(MigrationState *s)
 266 {
 267     int ret = 0;
 268
 269     if (s->file) {
 270         DPRINTF("closing file\n");
 271         ret = qemu_fclose(s->file);
 272         s->file = NULL;
 273     }
 274
 275     migrate_fd_close(s);
 276     return ret;
 277 }
 278
 279 void migrate_fd_error(MigrationState *s)
 280 {
 281     DPRINTF("setting error state\n");
 282     s->state = MIG_STATE_ERROR;
 283     notifier_list_notify(&migration_state_notifiers, s);
 284     migrate_fd_cleanup(s);
 285 }
 286
 287 static void migrate_fd_completed(MigrationState *s)
 288 {
 289     DPRINTF("setting completed state\n");
 290     if (migrate_fd_cleanup(s) < 0) {
 291         s->state = MIG_STATE_ERROR;
 292     } else {
 293         s->state = MIG_STATE_COMPLETED;
 294         runstate_set(RUN_STATE_POSTMIGRATE);
 295     }
 296     notifier_list_notify(&migration_state_notifiers, s);
 297 }
 298
 299 static void migrate_fd_put_notify(void *opaque)
 300 {
 301     MigrationState *s = opaque;
 302     int ret;
 303
 304     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 305     ret = qemu_file_put_notify(s->file);
 306     if (ret) {
 307         migrate_fd_error(s);
 308     }
 309 }
 310
 311 ssize_t migrate_fd_put_buffer(MigrationState *s, const void *data,
 312                               size_t size)
 313 {
 314     ssize_t ret;
 315
 316     if (s->state != MIG_STATE_ACTIVE) {
 317         return -EIO;
 318     }
 319
 320     do {
 321         ret = s->write(s, data, size);
 322     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 323
 324     if (ret == -1)
 325         ret = -(s->get_error(s));
 326
 327     if (ret == -EAGAIN) {
 328         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 329     }
 330
 331     return ret;
 332 }
 333
 334 void migrate_fd_put_ready(MigrationState *s)
 335 {
 336     int ret;
 337
 338     if (s->state != MIG_STATE_ACTIVE) {
 339         DPRINTF("put_ready returning because of non-active state\n");
 340         return;
 341     }
 342
 343     DPRINTF("iterate\n");
 344     ret = qemu_savevm_state_iterate(s->file);
 345     if (ret < 0) {
 346         migrate_fd_error(s);
 347     } else if (ret == 1) {
 348         int old_vm_running = runstate_is_running();
 349         int64_t start_time, end_time;
 350
 351         DPRINTF("done iterating\n");
 352         start_time = qemu_get_clock_ms(rt_clock);
 353         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 354         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 355
 356         if (qemu_savevm_state_complete(s->file) < 0) {
 357             migrate_fd_error(s);
 358         } else {
 359             migrate_fd_completed(s);
 360         }
 361         end_time = qemu_get_clock_ms(rt_clock);
 362         s->total_time = end_time - s->total_time;
 363         s->downtime = end_time - start_time;
 364         if (s->state != MIG_STATE_COMPLETED) {
 365             if (old_vm_running) {
 366                 vm_start();
 367             }
 368         }
 369     }
 370 }
 371
 372 static void migrate_fd_cancel(MigrationState *s)
 373 {
 374     if (s->state != MIG_STATE_ACTIVE)
 375         return;
 376
 377     DPRINTF("cancelling migration\n");
 378
 379     s->state = MIG_STATE_CANCELLED;
 380     notifier_list_notify(&migration_state_notifiers, s);
 381     qemu_savevm_state_cancel(s->file);
 382
 383     migrate_fd_cleanup(s);
 384 }
 385
 386 int migrate_fd_wait_for_unfreeze(MigrationState *s)
 387 {
 388     int ret;
 389
 390     DPRINTF("wait for unfreeze\n");
 391     if (s->state != MIG_STATE_ACTIVE)
 392         return -EINVAL;
 393
 394     do {
 395         fd_set wfds;
 396
 397         FD_ZERO(&wfds);
 398         FD_SET(s->fd, &wfds);
 399
 400         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 401     } while (ret == -1 && (s->get_error(s)) == EINTR);
 402
 403     if (ret == -1) {
 404         return -s->get_error(s);
 405     }
 406     return 0;
 407 }
 408
 409 int migrate_fd_close(MigrationState *s)
 410 {
 411     int rc = 0;
 412     if (s->fd != -1) {
 413         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 414         rc = s->close(s);
 415         s->fd = -1;
 416     }
 417     return rc;
 418 }
 419
 420 void add_migration_state_change_notifier(Notifier *notify)
 421 {
 422     notifier_list_add(&migration_state_notifiers, notify);
 423 }
 424
 425 void remove_migration_state_change_notifier(Notifier *notify)
 426 {
 427     notifier_remove(notify);
 428 }
 429
 430 bool migration_is_active(MigrationState *s)
 431 {
 432     return s->state == MIG_STATE_ACTIVE;
 433 }
 434
 435 bool migration_has_finished(MigrationState *s)
 436 {
 437     return s->state == MIG_STATE_COMPLETED;
 438 }
 439
 440 bool migration_has_failed(MigrationState *s)
 441 {
 442     return (s->state == MIG_STATE_CANCELLED ||
 443             s->state == MIG_STATE_ERROR);
 444 }
 445
 446 void migrate_fd_connect(MigrationState *s)
 447 {
 448     int ret;
 449
 450     s->state = MIG_STATE_ACTIVE;
 451     s->file = qemu_fopen_ops_buffered(s);
 452
 453     DPRINTF("beginning savevm\n");
 454     ret = qemu_savevm_state_begin(s->file, &s->params);
 455     if (ret < 0) {
 456         DPRINTF("failed, %d\n", ret);
 457         migrate_fd_error(s);
 458         return;
 459     }
 460     migrate_fd_put_ready(s);
 461 }
 462
 463 static MigrationState *migrate_init(const MigrationParams *params)
 464 {
 465     MigrationState *s = migrate_get_current();
 466     int64_t bandwidth_limit = s->bandwidth_limit;
 467     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 468     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 469
 470     memcpy(enabled_capabilities, s->enabled_capabilities,
 471            sizeof(enabled_capabilities));
 472
 473     memset(s, 0, sizeof(*s));
 474     s->bandwidth_limit = bandwidth_limit;
 475     s->params = *params;
 476     memcpy(s->enabled_capabilities, enabled_capabilities,
 477            sizeof(enabled_capabilities));
 478     s->xbzrle_cache_size = xbzrle_cache_size;
 479
 480     s->bandwidth_limit = bandwidth_limit;
 481     s->state = MIG_STATE_SETUP;
 482     s->total_time = qemu_get_clock_ms(rt_clock);
 483
 484     return s;
 485 }
 486
 487 static GSList *migration_blockers;
 488
 489 void migrate_add_blocker(Error *reason)
 490 {
 491     migration_blockers = g_slist_prepend(migration_blockers, reason);
 492 }
 493
 494 void migrate_del_blocker(Error *reason)
 495 {
 496     migration_blockers = g_slist_remove(migration_blockers, reason);
 497 }
 498
 499 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 500                  bool has_inc, bool inc, bool has_detach, bool detach,
 501                  Error **errp)
 502 {
 503     Error *local_err = NULL;
 504     MigrationState *s = migrate_get_current();
 505     MigrationParams params;
 506     const char *p;
 507
 508     params.blk = blk;
 509     params.shared = inc;
 510
 511     if (s->state == MIG_STATE_ACTIVE) {
 512         error_set(errp, QERR_MIGRATION_ACTIVE);
 513         return;
 514     }
 515
 516     if (qemu_savevm_state_blocked(errp)) {
 517         return;
 518     }
 519
 520     if (migration_blockers) {
 521         *errp = error_copy(migration_blockers->data);
 522         return;
 523     }
 524
 525     s = migrate_init(&params);
 526
 527     if (strstart(uri, "tcp:", &p)) {
 528         tcp_start_outgoing_migration(s, p, &local_err);
 529 #if !defined(WIN32)
 530     } else if (strstart(uri, "exec:", &p)) {
 531         exec_start_outgoing_migration(s, p, &local_err);
 532     } else if (strstart(uri, "unix:", &p)) {
 533         unix_start_outgoing_migration(s, p, &local_err);
 534     } else if (strstart(uri, "fd:", &p)) {
 535         fd_start_outgoing_migration(s, p, &local_err);
 536 #endif
 537     } else {
 538         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 539         return;
 540     }
 541
 542     if (local_err) {
 543         migrate_fd_error(s);
 544         error_propagate(errp, local_err);
 545         return;
 546     }
 547
 548     notifier_list_notify(&migration_state_notifiers, s);
 549 }
 550
 551 void qmp_migrate_cancel(Error **errp)
 552 {
 553     migrate_fd_cancel(migrate_get_current());
 554 }
 555
 556 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 557 {
 558     MigrationState *s = migrate_get_current();
 559
 560     /* Check for truncation */
 561     if (value != (size_t)value) {
 562         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 563                   "exceeding address space");
 564         return;
 565     }
 566
 567     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 568 }
 569
 570 int64_t qmp_query_migrate_cache_size(Error **errp)
 571 {
 572     return migrate_xbzrle_cache_size();
 573 }
 574
 575 void qmp_migrate_set_speed(int64_t value, Error **errp)
 576 {
 577     MigrationState *s;
 578
 579     if (value < 0) {
 580         value = 0;
 581     }
 582
 583     s = migrate_get_current();
 584     s->bandwidth_limit = value;
 585     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 586 }
 587
 588 void qmp_migrate_set_downtime(double value, Error **errp)
 589 {
 590     value *= 1e9;
 591     value = MAX(0, MIN(UINT64_MAX, value));
 592     max_downtime = (uint64_t)value;
 593 }
 594
 595 int migrate_use_xbzrle(void)
 596 {
 597     MigrationState *s;
 598
 599     s = migrate_get_current();
 600
 601     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 602 }
 603
 604 int64_t migrate_xbzrle_cache_size(void)
 605 {
 606     MigrationState *s;
 607
 608     s = migrate_get_current();
 609
 610     return s->xbzrle_cache_size;
 611 }